摘要
采用线性时不变系统把高维运动数据映射到低维状态空间;在低维状态空间中,定义了姿态之间的相似性度量;并采用误差平方和准则对时序的低维数据点集进行运动分割,分割点上的运动姿态被定义为关键帧.实验结果表明:该算法能够较好地提取出运动序列中的关键帧,并且这些关键帧能够很好地概括原始运动序列的内容.
Linear time-invariant system is used to derive an explicit mapping between the high- dimensional motion capture data and the low-dimensional state variables. A similarity metric is defined to measure the difference between different poses in the low-dimensional state space, and then the method of mean squared error is employed to divide the motion sequence into a sequence of concatenated segments. The poses at these segmentation points are then defined as keyframes. Experimental results show that the extracted keyframes by our method can give a good visual summarization of the original motion sequence.
出处
《计算机辅助设计与图形学学报》
EI
CSCD
北大核心
2008年第6期787-792,共6页
Journal of Computer-Aided Design & Computer Graphics
基金
国家自然科学基金(60573162,60533070,60603082)
国家“八六三”高技术研究发展计划(2006AA01Z336,2007AA01Z320)
中国科学院“科技助残行动计划”项目(KGCX2-YW-610)
关键词
运动捕获数据
关键帧
线性时不变系统
运动分割
motion capture data keyframe
linear time-invariant system
motion segmentation